基于平均熵值的多点地质统计建模参数优选方法

多点地质统计建模方法通过训练图像获取空间结构和相关性统计特征,重建结构复杂的储层地质模型,为提高油气预测效果服务。不同建模参数的选择会直接影响计算效率和模拟结果,因此选择合适的建模参数至关重要。鉴于传统参数灵敏度分析方法的不足,提出一种基于模式均熵的多点地质统计建模参数优选方法,采用Hsim相似度对随机模型和训练图像的模式均熵差异进行量化分析。以多点地质统计建模参数——样板尺寸为例,计算建模参数集和训练图像的空间及结构特征相似度,建立基于模式均熵差异的空间相关性评价指标和建模参数的拟合曲线,将相关性评价曲线趋于平稳的拐点所对应参数值作为最优参数。实验结果表明,相比传统参数优选方法,基于单点熵代替两点熵进行平均熵值计算的新方法可以准确客观地优选出多点地质统计建模算法的参数。

Parameter optimization method of multi-point geostatistical modeling based on average entropy

The multi-point geological statistical modeling method obtains the spatial structure and correlation statistical characteristics by training images, and reconstructs the reservoir geological model with complex structure, which serves to improve the effect of oil and gas prediction. The selection of different modeling parameters will directly affect the computational efficiency and simulation results, so it is very important to select the appropriate modeling parameters. In view of the deficiency of traditional parameter sensitivity analysis methods, a multi-point geostatistical modeling parameter optimization method based on pattern average entropy was proposed in this paper. Hsim similarity was used to quantitatively analyze the pattern entropy difference between random model and training image. Taking the multi-point geostatistical modeling parameter-template size as an example, the spatial and structural feature similarity between the modeling parameter set and the training image was calculated, and the spatial correlation evaluation index and the fitting curve of the modeling parameters based on the pattern mean entropy difference were established. the parameter values corresponding to the inflection point where the correlation evaluation curve tends to be stable were taken as the optimal parameters. Experimental results show that, compared with the traditional parameter optimization method, the new method based on single point entropy instead of two point entropy can accurately and objectively optimize the parameters of multi-point geostatistical modeling algorithm.